From: Big data hurdles in precision medicine and precision public health
Precision medicine | |||
 • Concentration on individualized treatment and neglect of time component of predictions, i.e. early risk vs. differential diagnosis vs. post-treatment survival  • Too much focus on genetics and –omics  • Research on actionable factors vs. immutable risk factors  • Integration of multi-omics  • Integration of multi-domain data (e.g. genetics, diet, lifestyle, social) | |||
Precision public health | |||
 • Definition of target units (e.g. ethnic groups, geographic zones, social groups)  • Conflict with precision medicine, i.e. individual-centric objectives (benefit of the single may not translate into benefit of the population)  • Population-level outcomes | |||
Data sources | Study designs | Prediction modelling | Translational relevance |
 • Heterogeneous data sources  • Unstructured data sources  • Lack of data on social determinants of health  • Measurement issues (e.g. incompleteness, inaccuracy, imprecision in self-reported data)  • Privacy and security  • Cost  • Limited adoption of common data models | • Semantic data integration (i.e. linking data elements by their meaning) • Large longitudinal cohorts • Ontology integration • Ontology appropriateness (e.g. ontologies made for billing vs. for diagnostic purposes) • Semantic interoperability • Automated study design | • Biases of all sorts (e.g. protopathic) • Confounding • Causal inference • Black-boxes vs. white-boxes (i.e. interpretability vs. performance) • Complexity-based model selection • Benchmark development • Pragmatic interoperability (reproducibility, replicability, generalizability) | • Limited individual empowerment • Disconnect from relevant clinical research • Personal health record/health avatar (besides provider’s electronic records) • Acceptance of artificial intelligence as integral part of doctors’ tools • Learning systems • Ethical usage and dissemination of modelling algorithms • Redefining disease phenotype |